search

wangkenpu / rsrgan

Robust Speech Recognition Using Generative Adversarial Networks (GAN)
MIT License
58 stars 16 forks source link

Some questions about datatset？ #1

Open shunfeichen opened 6 years ago

shunfeichen commented 6 years ago

Hi, I have read your paper about this repository. I did learn a lot. Now, I want to employ your code to make some experiments. I have some questions and hope you can give me some advice. (1) Do you use GAN in the condition( both reverberant and noisy)？ What's the results？ (2) Do I need to prepare a clean data before simulating the reverb data? However ， My data is collected form telephone and mobile phone.

wangkenpu commented 6 years ago

(1). Because of lacking noisy data, I didn't use GAN in both reverberant and noisy condition. But I think this architecture can work well in both reverberant and noisy condition. (2). Sure, I think you must prepare a clean dataset before simulating the reverberant data, because we need the ground truth during training. Although your data is colectted from telephone and mobile phone (maybe 8 kHz), if the data is not contaminated by noise and reverberation severely, you can regard it as clean data. Moreover, you acoustic model should be trained using these telephone and mobile phone data.

shunfeichen commented 6 years ago

Thank you very much!